Uncertainty quantification in ToxCast high throughput screening.

Area of Science:

• Toxicology
• Computational chemistry
• Data science

Background:

• High-throughput screening (HTS) generates vast chemical toxicity data, necessitating robust analysis methods.
• Existing models predict chemical toxicity but often lack thorough uncertainty quantification.
• Understanding uncertainty propagation is crucial for reliable HTS-based toxicity predictions.

Purpose of the Study:

• To explore the impact of parameter estimation uncertainties on HTS-derived toxicity predictions.
• To quantify uncertainty in model outputs using HTS data.
• To enhance the reliability of chemical toxicity predictions for risk assessment.

Main Methods:

• Nonparametric bootstrap resampling was used to calculate uncertainties in concentration-response parameters from HTS assays.
• The ToxCast estrogen receptor bioactivity model served as a case study for uncertainty propagation.
• Uncertainty in model outputs was quantified to assess prediction reliability.

Main Results:

• Uncertainty quantification identified potential false positives and false negatives in toxicity predictions.
• The distribution of model values around activity cutoffs was determined, increasing prediction confidence.
• High-variability chemical-assay results were flagged for manual review, optimizing expert resources.

Conclusions:

• This study demonstrates a robust method for uncertainty quantification in HTS data analysis.
• Improved confidence in HTS predictions facilitates their use in chemical risk assessment.
• The approach enhances the accuracy and reliability of predicting chemical bioactivity and toxicity.